Census Tract–Level Physical and Social Environmental Determinants of Edentulism: A Machine Learning Approach

TL;DR

This study uses machine learning to show how social and environmental factors at the local level affect tooth loss in older US adults.

Contribution

The novel use of XGBoost machine learning to analyze census tract-level data reveals how social and physical factors interact to influence edentulism.

Findings

The model explained 87% of the variance in edentulism rates among older adults using census tract data.

Social vulnerability and climate zone were the top predictors of edentulism.

Poverty and racial segregation in census tracts modified the impact of social and physical factors on edentulism risk.

Abstract

Edentulism (i.e., complete tooth loss) disproportionately affects older US adults in marginalized communities, driven by complex social and environmental factors. There is a critical gap in understanding how these factors interact at the population level to shape oral health disparities. This study employed Extreme Gradient Boosting (XGBoost), a machine learning technique, to analyze these determinants across 38,379 census tracts using 2021 data. Social environmental determinants—such as poverty rates, education levels, and employment status—were sourced from the American Community Survey and aggregated at the census tract level to capture localized socioeconomic conditions. Physical environmental determinants, including climate zone and access to dental care, were similarly aggregated to reflect tangible surroundings impacting health. The model predicted edentulism rates among adults aged 65+ with high precision, explaining 87% of the variance. Social vulnerability and climate zone emerged as the top predictors, underscoring the interplay between socioeconomic disadvantage and geographic conditions. Census tract characteristics, like poverty and racial segregation, moderated these effects, revealing distinct risk profiles—e.g., in high-poverty tracts, social factors amplified edentulism risk differently than in racially segregated ones. These findings highlight the contextual nature of oral health and demonstrate the value of localized, data-driven analyses. By identifying key social and physical determinants, this research informs targeted interventions to reduce oral health disparities. The use of XGBoost showcases machine learning’s potential to dissect complex health issues, offering actionable insights for policymakers and clinicians to improve outcomes in diverse settings.